Publications

Spatial transcriptomics assays are rapidly increasing in scale and complexity, making computational analysis a major bottleneck in biological discovery. Although frontier AI agents have improved dramatically at software engineering and general data analysis, it remains unclear whether they can extract biological insight from messy, real-world spatial datasets. We introduce SpatialBench, a benchmark of 146 verifiable problems derived from practical spatial analysis workflows spanning five spatial technologies and seven task categories. Each problem provides a snapshot of experimental data immediately prior to an analysis step and a deterministic grader that evaluates recovery of a key biological result. SpatialBench serves both as a measurement tool and a diagnostic lens for developing agents that can interact with real spatial datasets faithfully, transparently, and reproducibly.

Metagenomics has revolutionized our understanding of microbial communities, offering unprecedented insights into their genetic and functional diversity across Earth’s diverse ecosystems. Beyond their roles as environmental constituents, microbiomes act as symbionts, profoundly influencing the health and function of their host organisms. Given the inherent complexity of these communities and the diverse environments where they reside, the components of a metagenomics study must be carefully tailored to yield accurate results that are representative of the populations of interest. This Primer examines the methodological advancements and current practices that have shaped the field, from initial stages of sample collection and DNA extraction to the advanced bioinformatics tools employed for data analysis, with a particular focus on the profound impact of next-generation sequencing on the scale and accuracy of metagenomics studies. We critically assess the challenges and limitations inherent in metagenomics experimentation, available technologies and computational analysis methods. Beyond technical methodologies, we explore the application of metagenomics across various domains, including human health, agriculture and environmental monitoring. Looking ahead, we advocate for the development of more robust computational frameworks and enhanced interdisciplinary collaborations. This Primer serves as a comprehensive guide for advancing the precision and applicability of metagenomic studies, positioning them to address the complexities of microbial ecology and their broader implications for human health and environmental sustainability.

A common task in the analysis of microbial communities involves assigning taxonomic labels to the sequences derived from organisms found in the communities. Frequently, such labels are assigned using machine learning algorithms that are trained to recognize individual taxonomic groups based on training data sets that comprise sequences with known taxonomic labels. Ideally, the training data should rely on labels that are experimentally verified-formal taxonomic labels require knowledge of physical and biochemical properties of organisms that cannot be directly inferred from sequence alone. However, the labels associated with sequences in biological databases are most commonly computational predictions which themselves may rely on computationally-generated data-a process commonly referred to as "transitive annotation". In this manuscript we explore the implications of training a machine learning classifier (the Ribosomal Database Project’s Bayesian classifier in our case) on data that itself has been computationally generated. We demonstrate that even a few computationally-generated training data points can significantly skew the output of the classifier to the point where entire regions of the taxonomic space can be disturbed. We also discuss key factors that affect the resilience of classifiers to transitively-annotated training data, and propose best practices to avoid the artifacts described in our paper.

Sequence differences between the strains of bacteria comprising host-associated and environmental microbiota may play a role in community assembly and influence the resilience of microbial communities to disturbances. Tools for characterizing strain-level variation within microbial communities, however, are limited in scope, focusing on just single nucleotide polymorphisms, or relying on reference-based analyses that miss complex functional and structural variants. Here, we demonstrate the power of assembly graph analysis to detect and characterize structural variants in almost 1,000 metagenomes generated as part of the Human Microbiome Project. We identify over nine million variants comprising insertion/deletion events, repeat copy-number changes, and mobile elements such as plasmids. We highlight some of the potential functional roles of these genomic changes. Our analysis revealed striking differences in the rate of variation across body sites, highlighting niche-specific mechanisms of bacterial adaptation. The structural variants we detect also include potentially novel prophage integration events, highlighting the potential use of graph-based analyses for phage discovery.

For decades, the 16S rRNA gene has been used to taxonomically classify prokaryotic species and to taxonomically profile microbial communities. However, the 16S rRNA gene has been criticized for being too conserved to differentiate between distinct species. We argue that the inability to differentiate between species is not a unique feature of the 16S rRNA gene. Rather, we observe the gradual loss of species-level resolution for other nearly-universal prokaryotic marker genes as the number of gene sequences increases in reference databases. This trend was strongly correlated with how represented a taxonomic group was in the database and indicates that, at the gene-level, the boundaries between many species might be fuzzy. Through our study, we argue that any approach that relies on a single marker to distinguish bacterial taxa is fraught even if some markers appear to be discriminative in current databases.

Sequence differences between the strains of bacteria comprising host-associated and environmental microbiota may play a role in community assembly and influence the resilience of microbial communities to disturbances. Tools for characterizing strain-level variation within microbial communities, however, are limited in scope, focusing on just single nucleotide poly-morphisms, or relying on reference-based analyses that miss complex structural variants. In this thesis, we describe data-driven methods to characterize variation in metagenomic data.In the first part of the thesis, I present our analysis of the structural variants identified from metagenomic whole genome shotgun sequencing data. I begin by describing the power of assembly graph analysis to detect over 9 million structural variants such as insertion/deletion, repeat copy-number changes, and mobile elements, in nearly 1,000 metagenomes generated as a part of the Human Microbiome Project. Next, I describe Binnacle, which is a structural variant aware binning algorithm. To improve upon the fragmented nature of assemblies, metagenomic binning is performed to cluster contigs that is likely to have originated from the same genome. We show that binning “graph-based” scaffolds, rather than contigs, improves the quality of the bins, and captures a broader set of the genes of the genomes being reconstructed. Finally, we present a case study of the microbial mats from the Yellowstone National Park. The cyanobacterium Synechococcus is abundant in these mats along a stable temperature gradient from ∼ 50°C to ∼ 65°C and plays a key role in fixing carbon and nitrogen. Previous studies have isolated and generated good quality reference sequences of two major Synechococcus spp. that share a very high genomic content; OS-A and OS-B’. Despite the high abundance of the Synechococcus spp., metagenomic assembly of these organisms is challenging due to the large number of rearrangements between them. We explore the genomic diversity of the Synechococcus spp. using a reference genome, reliant assembly and scaffolding. We also highlight that the variants we detect can be used to fingerprint the local biogeography of the hot spring.In the second part of the thesis, I present our analysis of amplicon sequencing data, specifically the 16S rRNA gene sequences. I begin by describing SCRAPT (Sample, Cluster, Recruit, AdaPt and iTerate), which is a fast iterative algorithm for clustering large 16S rRNA gene datasets. We also show that SCRAPT is able to produce operational taxonomic units that are less fragmented than popular tools: UCLUST, CD-HIT and DNACLUST and runs orders of magnitude faster than existing methods. Finally, we study the impact of transitive annotation on taxonomic classifiers. Taxonomic labels are assigned using machine learning algorithms trained to recognize individual taxonomic groups based on training sequences with known taxonomic labels. Ideally, the training data should rely on experimentally verified-formal taxonomic labels however, the labels associated with sequences in biological databases are most commonly the result of computational predictions– “transitive annotation.” We demonstrate that even a few computationally-generated training data points can significantly skew the output of the classifier to the point where entire regions of the taxonomic space can be disturbed. We also discuss key factors that affect the resilience of classifiers to transitively annotated training data and propose best practices to avoid the artifacts described in this thesis.

16S rRNA gene sequence clustering is an important tool in characterizing the diversity of microbial communities. As 16S rRNA gene data sets are growing in size, existing sequence clustering algorithms increasingly become an analytical bottleneck. Part of this bottleneck is due to the substantial computational cost expended on small clusters and singleton sequences. We propose an iterative sampling-based 16S rRNA gene sequence clustering approach that targets the largest clusters in the data set, allowing users to stop the clustering process when sufficient clusters are available for the specific analysis being targeted. We describe a probabilistic analysis of the iterative clustering process that supports the intuition that the clustering process identifies the larger clusters in the data set first. Using real data sets of 16S rRNA gene sequences, we show that the iterative algorithm, coupled with an adaptive sampling process and a mode-shifting strategy for identifying cluster representatives, substantially speeds up the clustering process while being effective at capturing the large clusters in the data set. The experiments also show that SCRAPT (Sample, Cluster, Recruit, AdaPt and iTerate) is able to produce operational taxonomic units that are less fragmented than popular tools: UCLUST, CD-HIT and DNACLUST. The algorithm is implemented in the open-source package SCRAPT. The source code used to generate the results presented in this paper is available at https://github.com/hsmurali/SCRAPT.

Bacterial species often undergo rampant recombination yet maintain cohesive genomic identity. Ecological differences can generate recombination barriers between species and sustain genomic clusters in the short term. But can these forces prevent genomic mixing during long-term coevolution? Cyanobacteria in Yellowstone hot springs comprise several diverse species that have coevolved for hundreds of thousands of years, providing a rare natural experiment. By analyzing more than 300 single-cell genomes, we show that despite each species forming a distinct genomic cluster, much of the diversity within species is the result of hybridization driven by selection, which has mixed their ancestral genotypes. This widespread mixing is contrary to the prevailing view that ecological barriers can maintain cohesive bacterial species and highlights the importance of hybridization as a source of genomic diversity.

High-throughput sequencing has revolutionized the field of microbiology, however, reconstructing complete genomes of organisms from whole metagenomic shotgun sequencing data remains a challenge. Recovered genomes are often highly fragmented, due to uneven abundances of organisms, repeats within and across genomes, sequencing errors, and strain-level variation. To address the fragmented nature of metagenomic assemblies, scientists rely on a process called binning, which clusters together contigs inferred to originate from the same organism. Existing binning algorithms use oligonucleotide frequencies and contig abundance (coverage) within and across samples to group together contigs from the same organism. However, these algorithms often miss short contigs and contigs from regions with unusual coverage or DNA composition characteristics, such as mobile elements. Here, we propose that information from assembly graphs can assist current strategies for metagenomic binning. We use MetaCarvel, a metagenomic scaffolding tool, to construct assembly graphs where contigs are nodes and edges are inferred based on paired-end reads. We developed a tool, Binnacle, that extracts information from the assembly graphs and clusters scaffolds into comprehensive bins. Binnacle also provides wrapper scripts to integrate with existing binning methods. The Binnacle pipeline can be found on GitHub (https://github.com/marbl/binnacle). We show that binning graph-based scaffolds, rather than contigs, improves the contiguity and quality of the resulting bins, and captures a broader set of the genes of the organisms being reconstructed.

During demand response events, utility customers typically make decisions to reduce their own load independent of other customers. In the presence of district cooling with a centralized cooling plant, the control decisions of buildings and the district cooling plant become coupled, since the setpoints chosen by the cooling plant affect the consumption of the buildings and vice versa.While past works on demand response address control of building-level heating and cooling systems, control of such systems in the presence of district cooling has not received much attention. We consider the problem of minimizing the discomfort of buildings while meeting the target demand reduction. Specifically, we identify the optimal setpoints for the buildings and district cooling plant, even while obeying the non-linear, coupled thermodynamic constraints between the district cooling plant and buildings. We propose a novel solution strategy using domain knowledge that transforms the complex non-linear optimization problem to a series of quadratic programming which can be then solved conventionally.We validate the performance of the proposed strategy by comparing it with a combinatorial brute-force solution on a small data set. We also evaluate the performance of our strategy on a real-world data set of 416 buildings that are served by a district cooling plant. The results indicate that including the district plant in demand response and solving with the coupled constraints, allows the utility to meet higher target reductions for the same comfort levels. Also, the proposed solution strategy is both fast (at least 4x) and scalable (35x) when compared with conventional optimization solvers.